Sheet 05 of 08AI chatbots in production

Three chatbots, real data

Three chatbots in production, each answering from real business data: a multi-agent chatbot on Google ADK that lets Aepsy providers ask about their bookings and appointments; an analytics chatbot over the product tracking data, answering about marketing performance, funnels and conversion rates with charts, built with the data and AI engineers; and an internal analytics chatbot at GEN answering about usage, jobs, AI API spend, error rates and user feedback on generated results. In my own product the AI sits behind a server-side layer, so a provider change never reaches the client.

Period: Sep 2025 - PresentGEN, Aepsy
Role: Multi-agent and analytics chatbotsAepsy and GEN
Stack: Google ADK, LLM tool calling
Proof

Agent and tools

Fig. 1

Fig. 1A provider's question goes to the agent on Google ADK, which calls a tool for bookings and appointments and answers from that data.

Inventory

Fig. 2

Fig. 2Three chatbots in production, each answering from real business data, and the provider layer in my own product.

Facts

01 to 05

Multi-agent chatbot on Google ADK for providers' bookings and appointments.
From my CV, Aepsy
Analytics chatbot over product tracking data: marketing performance, funnels, conversion rates, with charts; built with the data and AI engineers.
From my CV, Aepsy
An internal analytics chatbot on the multi-tenant platform answering about usage, jobs, AI API spend, error rates and user feedback on generated results.
From my CV, GEN
Giáo Án AI: a server-side layer over the AI providers (a provider change never reaches the client).
From my CV, Giáo Án AI
Google ADK multi-agent systems, Claude API, LLM tool calling, streaming chat UIs, MCP.
From my CV, skills

Why it is hard

A chatbot over business data fails in ways a demo never shows, and grounding its answers in bookings, funnels and spend is the actual work.